The recorded _sweep_limit may not necessarily remain a block boundary as the old generation expands during a concurrent cycle. Terminal actions inside the sweep closure need to be aware of this as they cross over the limit.
Reviewed-by: johnc, minqi
7002546: regression on SpecJbb2005 on 7b118 comparing to 7b117 on small heaps
Relaxed assertion checking related to incremental_collection_failed flag to allow for ExplicitGCInvokesConcurrent behaviour where we do not want a failing scavenge to bail to a stop-world collection. Parameterized incremental_collection_will_fail() so we can selectively use, or not use, as appropriate, the statistical prediction at specific use sites. This essentially reverts the scavenge bail-out logic to what it was prior to some recent changes that had inadvertently started using the statistical prediction which can be noisy in the presence of bursty loads. Added some associated verbose non-product debugging messages.
Reviewed-by: johnc, tonyp
Deprecated HandlePromotionFailure, removing the ability to turn off that feature, did away with one epoch look-ahead when deciding if a scavenge is likely to fail, relying on current data.
Reviewed-by: jmasa, johnc, poonam
Inserted missing yield(check)s in closures used during the work-stealing phase of parallel concurrent marking, a missing synchronous yield-request in the cms perm gen allocation path, and a terminator-terminator for the offer_termination invocation that monitors the yield status of the concurrent marking task. Elaborated some documentation comments and made some task queue termination loop flags configurable at start-up to aid debugging in the field.
Reviewed-by: jmasa, johnc, poonam
Allow for the possibility (when the heap is expanding) that the sweep might skip over and past, rather than necessarily step on, the sweep limit determined at the beginning of a concurrent marking cycle.
Reviewed-by: jmasa, tonyp
GC workers now recognize an intermediate transient state of blocks which are allocated but have not yet completed initialization. blk_start() calls do not attempt to determine the size of a block in the transient state, rather waiting for the block to become initialized so that it is safe to query its size. Audited and ensured the order of initialization of object fields (klass, free bit and size) to respect block state transition protocol. Also included some new assertion checking code enabled in debug mode.
Reviewed-by: chrisphi, johnc, poonam
Moved PromotionInfo and friends into new files promotionInfo.{h,c}pp from their previous compactibleFreeListSpace.{h,c}pp home.
Reviewed-by: apetrusenko
On sun4v/CMT avoid use of memset() in BOT updates so as to prevent concurrent BOT readers from seeing the phantom zeros arising from memset()'s use of BIS.
Reviewed-by: jmasa, johnc, minqi, poonam, tonyp
Autonomic per-worker free block cache sizing, tunable coalition policies, fixes to per-size block statistics, retuned gain and bandwidth of some feedback loop filters to allow quicker reactivity to abrupt changes in ambient demand, and other heuristics to reduce fragmentation of the CMS old gen. Also tightened some assertions, including those related to locking.
Reviewed-by: jmasa
Treat ProfileData in MDO's as a source of weak, not strong, roots. Fixes the bug for stop-world collection -- the case of concurrent collection will be fixed separately.
Reviewed-by: jcoomes, jmasa, kvn, never
When using compressed oops, rather than chaining the overflowed grey objects' pre-images through their klass words, we use GC-worker thread-local overflow stacks.
Reviewed-by: jcoomes, jmasa
Avoid overflow list walk in CMS & ParNew when it is unnecessary. Fix a couple of correctness issues, including a C-heap leak, in ParNew at the intersection of promotion failure, work queue overflow and object array chunking. Add stress testing option and related assertion checking.
Reviewed-by: jmasa
For current soft-ref clearing policies, we can decide at marking time if a soft-reference will definitely not be cleared, postponing the decision of whether it will definitely be cleared to the final reference processing phase. This can be especially beneficial in the case of concurrent collectors where the marking is usually concurrent but reference processing is usually not.
Reviewed-by: jmasa
When we encounter marking stack overflow during precleaning of Reference lists, we were using the overflow list mechanism, which can cause problems on account of mutating the mark word of the header because of conflicts with mutator accesses and updates of that field. Instead we should use the usual mechanism for overflow handling in concurrent phases, namely dirtying of the card on which the overflowed object lies. Since precleaning effectively does a form of discovered list processing, albeit with discovery enabled, we needed to adjust some code to be correct in the face of interleaved processing and discovery.
Reviewed-by: apetrusenko, jcoomes
Fixed CMSConcMarkingTask::reset() to store the restart address upon a marking stack overflow and to use it as the base, suitably aligned, for restarting the scan in CMSConcMarkingTask::do_scan_and_mark().
Reviewed-by: jcoomes, tonyp
When an object array overflows during precleaning, we should have been marking the entire array dirty, not just its first card.
Reviewed-by: jmasa, poonam, tonyp
